Abstract: We present a novel skeleton-based approach for performing shape correspondence by exploiting the mesh geometric properties associated with the skeletons of the objects. The skeletal nodes which correspond to the similar parts of the two objects are matched. We compute a set of potential shape correspondences and then evaluate the best one. Our method may merge adjacent skeletal nodes and adjacent skeletal branches so that it can compute not only an 1-1 correspondence but also a many-many correspondence. We apply our method to various models with similar topological structures and compare with the latest techniques. Experimental results show that our method can compute acceptable shape correspondence.